weekly recap #2: apr 27 - may 3, 2026

i have been making an effort to read higher quality things more often. the reason being is that i strongly feel that good inputs = good outputs. there are people who spend weeks and months (and even years) working on a problem, and then talk about it in detail online. to me it's unfathomable that this is available to me. that someone spent the time and effort working on the hard problem, and then just publishes the experience and the wisdom they gained from it for anyone to read and extract for themselves.

i think it goes back to this idea that we stand on the shoulders of giants. most new ideas do not come from 0. they come from building on top of someone else's new idea. at the most atomic level (computers) or highest (payment rails). this makes it much easier to think of things that are novel. we can learn from other people's mistakes instead of having to test it out on our own.

a common mistake with this is that people realize this, but narrow their thinking when it comes to what they can extract experience from. for example, thinking it has to be within their own domain for it to be useful. but that is not true. you can make connections to things that are in the farthest domains possible. for example, Henry Ford invented the assembly line for his cars by observing meatpacking disassembly lines in chicago.

to me it feels similar to the quote "there are cathedrals everywhere for those with the eyes to see"

what i read this week

• was feeling a bit sick this week, so did not include alot of commentary around some of the stuff read this week. next week will be better hopefully!
• tools like amp, droid, pi building on top of the anthropic platform was initially the goal with the company itself, but now a tool like claude code (that was initially created to get people to use the platform more) is competing with these tools in an almost destructive way. even though it's beneficial for these other tools to succeed.
  • tweet
• writing anything online is pretty hard. what prevents me is that what i want to write is not useful or helpful for people. i was reading this, and it gave me a different perspective: what prevents some people is the fact that they don't want to offend people with contrarian takes. that is a bit surprising, but interesting

  • An idea I'm trying to hold tighter is that the only way to be useful here is to risk offending people. And perhaps if you're not taking any risk offending people then it's unlikely you're saying something useful.

• nice explainer on gas refunds in evm
• cool dashboard by the everclear team showing the various fees for important rebalancing routes
• ghostty is leaving github
• known costs unknowable benefits
• Raising Your Ceiling
• this is one of the best things i have read in a long time. it completely changed my perspective on the concept of having kids (as was foretold in the intro lol). mainly, the attitude section changed it. especially the part about:

  • If you adopt the right attitude, most of the challenges of having kids are a sort of "amusing struggle" where they're psychologically taxing in the moment but also funny, like trying to get a drunk person to drink water before bed.

• apparently garden leave is a thing, although not surprised
• this is just a mess Litecoin MWEB Security Incident Postmortem
• risking joy. "Snowdrops" by Louise Gluck
• nothing will change your life more than saying what you actually want to say
• stripe built a site where you input your API url and their agent will review it to see if it follows their design principles.
• list of all of the exploits in crypto this past month (april). $624 million exploited total.
• write screenshot essays
• good bios

  • PREVIOUSLY MANY THINGS | CURRENTLY NEW THINGS

• openai guide on how to prompt gpt-5.5
  • it's fascinating that gpt-5.5 and opus-4.7 have similar instructions to stop trying to define the process to solve the problem in the prompt and instead let the model find the most efficient path. prompting should also be much more outcome-focused and well defined. i have been experiencing this with opus-4.7 compared to 4.6, where with 4.6 it would assume my desired outcome if i left out details but 4.7 asks for clarity on its own.
• since GPT-5.1, GPT models have had a strange affinity to talk about goblins unprompted. OpenAI released a deep dive into why, really cool read!
• i've been trying to get better at chess the past couple months, nice little blog on someone who climbed to 2k!
• linear seems to be going after github. linear releases & git diff within linear now
• git-sync, can't think of any personal use cases but certainly useful
• Warp went open source a couple days ago.
  • it's cool that they are also now releasing deep dives on how certain things work, like a PTY
• the lichess blog is underrated
  • where do grandmasters play chess?
  • woodpecker method training
• treat agent output like compiler output

  • this above post is summarized just from this paragraph:

    The compiler analogy is clarifying here not because it tells us to trust agents blindly, but because it shows us what a mature pipeline looks like once you stop treating artifacts as things to be read and start treating them as things to be verified. We don't review compiled binaries. We run them against test suites, we check them against specifications, we instrument them in production.

• related to the treating agents output like compiler output post
• scrolling is on the decline.
• solomon's paradox
• growth engineering. always a fan of data based decision making instead of subjectivity.